Any plans to create feature flag code management project?

[cosmic-flood]

Hi,

As more teams are transitioning their code from their original third-party or custom SDKs to the standardized OpenFeature SDK, I believe it's a good time to initiate code reference and stale feature flag management community-level project. This project could assist developers in managing their feature flags in their code with their preferred code editors.

I'm curious to know if OpenFeature has a stance on this.

Thank you.

[beeme1mr]

Hey @cosmic-flood, could you please describe what you had in mind? A full-fledged feature management system is out of scope for OpenFeature as we want to make it easy for people to use the management tool of their choice. However, there may be opportunities to build tooling that works across providers to help support best practices.

[cosmic-flood]

I recently experimented with using ChatGPT to identify and remove dead feature flags in a project. The results were quite promising. However, to effectively utilize ChatGPT, you need to prepare context-specific data for each feature flag provider, like LaunchDarkly. This preparation can be quite labor-intensive.

Now, as people begin migrating their SDKs from provider-specific ones to those based on OpenFeature, this process of preparing data could become easier and more standardized.

This means that if ChatGPT, or tools like Copilot, continue to reduce operational costs, we might not need to develop a specialized tool. Instead, we could leverage the standardization offered by OpenFeature.

I've written a blog post about how to eliminate dead feature flags using ChatGPT prompts, which can be found at FeatBit Blog.

[beeme1mr]

Using ChatGPT to help clean up stale feature flags sounds like an interesting idea to me. How do you think OpenFeature can help? Are you looking for prompt templates or some other way to train the LLM? Perhaps a blog similar to the one you wrote fro FeatBit may be a good way to get people thinking about the potential.

[cosmic-flood]

I don't believe any LLM other than ChatGPT-4 or Bard, can achieve this goal, as true AGI reasoning power is essential. In my understanding of AI models, a large, deep learning network with a significant volume of parameters is necessary to build its Bayesian/Attention mechanism reasoning capabilities.

I think I'll continue my exploration, possibly with Copilot.

[beeme1mr]

Please let me know what you find. It will be increasingly important to ensure our documentation works well with LLMs.